Trading bitcoin profitably remains more of an art than an exact science.
That may change, however, with a new paper that claims to have devised a trading strategy that can produce an 89% return in less than two months.
The authors, Massachusetts Institute of Technology associate professor Devavrat Shah and computer science student Kang Zhang, collected data from OKCoin, the world’s largest exchange by trading volume, from February to July.
They fed the data into a predictive statistical model they have developed and used the results to conduct a simulation of CNY/BTC trades. In the simulation, the trader could only go long or short 1 BTC in each trade.
Volatility boosts profits
The trading simulation, conducted on data taken from 50 consecutive days in May and June, produced highly profitable results. The simulated trader invested 3,781 yuan and made 2,872 trades. The total cumulative profit was 3,362 yuan, or an 89% return on the amount invested.
The trading strategy produced the greatest profits when volatility was high, in the period at the end of May and the start of June, and was still profitable when the price declined steadily at the end of the simulated period.
The trading strategy also produced a Sharpe ratio of 4.1, the authors write. This expresses a portfolio’s return after adjusting for the risk-free rate of return. A high ratio shows that an investor produced returns while taking on less risk, with scores of three and over being considered excellent.
The authors’ Sharpe ratio compares favourably to benchmark mutual funds, like the Vanguard Total Stock Market Index Fund, the world’s largest such vehicle, which is worth $355bn. That fund has a one-year Sharpe ratio of 1.79 and has returned 8.32% in the last year.
Peering into the data
The paper’s results may also support the claims of technical traders in bitcoin markets. The authors analysed their prediction data and found evidence of ‘triangle’ and ‘head and shoulders’ patterns in the price charts.
“This seems to suggest that there are indeed such patterns and [… explains] the success of our trading strategy,” they write.
A preliminary version of the paper, titled Bayesian Regression and Bitcoin, was published in the Proceedings of the 2014 Allerton Conference on Communication, Control and Computing – one of the longest-running and most prestigious conferences in its field. The three-day conference concluded on 3rd October.
Given the simulation’s restricted trade size of 1 BTC, could more money be made with more capital at stake? The authors write that more research is required, although they speculate that profit can be magnified.
The authors also note that further profits could be produced by crunching more data, although this would require “computation at a massive scale”. They used a 32-core machine with 128GB of RAM for the study and “representative” time-series data at the predictive modelling stage.
Origins in Twitter analysis
Shah and Zhang’s predictive simulation is based on a ‘latent source model’ that was described in a paper published last year and was designed to predict what would become ‘trending topics’ on Twitter.
Shah co-authored that paper with two researchers at MIT and Twitter. Their model was able to predict trending topics accurately up to 79% of the time, according to the authors.
Profits image via Shutterstock